Tableau AI: My 2024 Retrospective on Tableau Agents, Agentic Ai and Salesforce

0:00This video is going to be about Tableau and AI.

0:03Now I tried to make this video in late 2024 and it was largely because there was something about Tableau's AI efforts that wasn't quite connecting with the way that I was using the product.

0:14and I couldn't quite put a finger on it and I made this video three times and each time I just couldn't come to terms with what I'd made.

0:21There was something missing.

0:22I just didn't understand what it was though.

0:24So I decided to use AI a bit more for a lot longer.

0:28and think more critically about it and go on a journey of discovery mainly to understand what it was that I didn't like about some AI features.

0:36over other features.

0:37Not just in Tableau, this is across the whole industry.

0:40And so in this video I've pulled all of that into one context to try and create a marker in the sand so we can look at this video as a point in time of how Tableau was doing with AI.

0:50Now the other thing is if you're at Salesforce and you're watching this, you know more than anyone that I'm the biggest fan of the product.

0:56But I think this video is a necessary part of continually challenging you to make a better product

1:02And so I'm really open to constructive feedback.

1:05A lot of you reach out to me on Slack and give me sort of uh points in return to what I've said here in the video.

1:10Sometimes you even drop comments below.

1:12So that's very much your stage

1:14here and on LinkedIn to sort of get back to me as well.

1:16Okay, the context is set, the barrel's loaded, let's fire some shots.

1:20You know what comes next?

1:22Let's get stuck in.

1:23You'll see from the timestamps in this video that in order to be able to really kind of have this discussion in the constructive way

1:30We must first look at the industry in general, AI specifically, and uh narrow in on data and analytics in that industry to really set the standard upon which Salesforce and By Extension Tableau

1:41can be judged against.

1:42And so what I wanted to do is just bring everyone up to speed on what these big trends are.

1:46If you already know what happened in 2024, just skip ahead to the next chapter.

1:49But I think it's important to touch on this because this is the context upon which I'll be judging Tableau and their capabilities.

1:56So 2024 was extremely confusing when it comes to AI because it just felt like throughout the whole year the marketing teams at some of the largest companies in the world were two pages ahead.

2:06of the engineers who are actually building the features.

2:08So you'd often get the marketing well before the feature arrived.

2:11And so you had this sort of exchange where

2:14Companies would promise these really big promises, but then they'd sort of fall short when they actually came to pass.

2:19And we were also even meant to get specific models released this year, but they didn't really transpire.

2:24But

2:24If you were paying attention, there were really a handful of key players that you'll probably know about, you've heard about.

2:30The first one is obviously OpenAI and ChatGPT.

2:32I'm just looking at my notes here.

2:34The big thing that ChatGPT did is it really increased the

2:37uh range and capability of all their models, not just doing the standard models that we've been used to in Chat GPT, but taking that to a a future of multimodal access.

2:46So being able to work with images, text and video

2:49all in the same model but then also releasing different models that you could use specifically for creating videos and even reasoning in the most recent example just before Christmas

2:59The other big player was Google with Gemini.

3:02Now Gemini is really promising because Google's really pushing forward when it comes to context.

3:06They have some of the largest windows for context.

3:09By context, we're talking about the amount of information you can provide, the AI model

3:13in order for it to process your request.

3:15So Google are really pushing forward with the context they want to bring to you.

3:18And also their latest model, Gemini 3.

3:212, is actually one of the most powerful models.

3:23allowing you to live stream capabilities to the model.

3:27Andy Cockgrove made a good video recently showing you how you could do this to essentially replace me in terms of Tableau tutorial.

3:33So go ahead and check out that video.

3:34I'll put a link in the description.

3:36The other player that I think a lot of people are sleeping on is Anthropic with Claude.

3:40Now Claude is very similar to ChatGPT, but I actually find it much, much more capable.

3:44In fact, throughout the whole of 2024, I only used Claude for pretty much all my coding tasks.

3:49And I typically demoted ChatGPT to frivolous use, just being able to ask a question quickly on an app.

3:55ChatGPT is farther ahead because they have the experience dialed in.

3:58But when Claude catches up, I'll mostly be using Claude.

4:01And you'll find that Claude also have a much better capability of understanding context.

4:06And I just found some of the outputs that you know Claude was putting out just felt more reasonable.

4:10They felt more workable, much easier to edit, much more pleasant to read.

4:14The last company to mention here is obviously Meta.

4:16Their big push is making AI models more open source.

4:19Essentially they have a licensing structure that stops the really large companies from capitalizing on their model

4:24but allows everyone else to run these models on their laptops.

4:27I actually have a working version of the latest Lama model uh in the largest form running on my laptop, which is absolutely wild to think that that's

4:35kind of possible without having to send my data to the cloud I think is incredible.

4:39So those were the four big players.

4:41Now the one player that I haven't mentioned yet is NVIDIA

4:44NVIDIA is notable here because they make all the hardware that many of these companies are using to not only run these models but also build models.

4:51So if you're

4:52Paying attention to the AR race in 2024, you'll know that NVIDIA is making the hardware and making the sort of foundational leaps in terms of architecture and infrastructure to allow companies to actually build models at scale.

5:04But that said, every other company who makes hardware is very close behind trying to eat into that market.

5:11And so with the big companies and players out of the way, it's now really easy to sort of synthesize

5:15what the big trends were.

5:17And for the record, many people will disagree on this.

5:19I've just pulled out what I think were the salient points from the year.

5:22But the first one is multimodal design, essentially models that can take in multiple forms of input

5:27to process a request we've already touched on that with chat GPT but actually Google are doing the same thing and I think in the future all models will have multimodal context

5:35The other one more subtle but linked to multimodal input is giving models larger context windows.

5:41So this is something that I mentioned Google is good at

5:43That is really important because it allows us to put larger and larger datasets to the point where we can actually start to answer meaningful questions when it comes to business

5:51For example, if I wanted to take an entire PDF and ask an LLM to process it and find specific bits of information, previously that was pretty hard in 2023.

6:01Now the context windows are large enough in most models for you to be able to put large documents, even books, inside of the context window and then ask the model lots of different

6:09from questions.

6:10We'll come back to this in a short while.

6:11The other one you will have heard from Salesforce specifically is agentic AI.

6:15Now agentic AI seems to be like a an outcome of the industry realizing that chat models and LLMs were basically not that useful for business context.

6:24What people were really looking for is autonomous solutions.

6:27And so they've sort of crystallized around this concept of agentic AI, this idea that there are lots of autonomous agents.

6:34doing tasks on behalf of users, saving time, increasing productivity, and potentially doing things that are actually quite hard to do in a manual way or in a direct way.

6:43So this is something really big.

6:44You'll have seen Salesforce talk about agent force, their push into this specific field, and we'll come back to that later because obviously we'll dig into Tableau and Salesforce

6:52specifically.

6:53The next one is edge computing.

6:55This is a very simple concept.

6:56At the moment all AR models you run them in the browser or if you've got meta locally on your machine.

7:01Now being able to run it locally on your machine is actually quite useful because it saves you having to send data to the cloud.

7:06And this is what essentially edge computing is.

7:09Essentially these large companies making their models small enough so they can be fit into a local context.

7:15The local context being your data center, your business.

7:18And it allows you to run these models locally.

7:20So very simple example, maybe I've got a CCTV camera in front of my home.

7:24Instead of hooking it up to the cloud and streaming my video over to Google,

7:28Instead with edge computing I can buy a computer, put it in a network in my house and just have the doorbell camera stream that input right to that specific node.

7:38So that's much much easier and it's actually quite advantageous

7:41In specific instances where you need the sort of data and information to happen really quite fast.

7:47The downside with the internet is that it takes time to send information back and forth.

7:51Edge computing allows this processing to happen locally and it allows it to happen much faster.

7:56The other one was the proliferation of advanced AI architecture.

8:00Specifically RAG Retrieval Augmented Generation and the other one is COT Chain of Thought.

8:06Now with RAG, the most notable example here is Google's notebook LLM

8:10Now this is quite an interesting one because what it allows you to do is upload a bunch of documents and then it can do things like generate a podcast based on those documents.

8:18But the idea here is that

8:19you're grounding your AI model in the information that you give it, which makes it much, much easier to do things like I could upload a bunch of insurance document and then ask the model a bunch of questions about all of those documents.

8:31But then the key thing is that the AR models is able to link to specific context, specific pages, and even highlight information for us, which makes it much easier to pass.

8:42large sets of information without needing to pass them which is absolutely brilliant.

8:47The other one you'll have seen just before Christmas is Chain of Thought.

8:50So this is all about AI models starting to behave like humans, essentially showing their reasoning and processing problems in small chunks, but then showing you their working as they go along.

8:59This is a very advanced use case, but I think this is a very prominent type of AR model that we'll see a lot in 2025.

9:05And the last one I've already mentioned, but this is the concept of foundational models.

9:09Essentially taking any of these models

9:11packaging them up and then building enhanced capabilities on top of them to do very very specific things and grinding those concepts so that essentially your model can be trained to do one task very very well

9:21And so those were the large trends.

9:23I could make a documentary about this if I had the time and I knew more about it, but I don't, so I'm gonna stop it here.

9:28I think if you've got any points that you think I've missed, put them in the comments below

9:31But I think these are the main trends that we want to take forward into our understanding of how Tableau is performed with AI, and we can start to see how Salesforce has adopted these methods.

9:41Okay, now that you've covered the trends, you'll obviously start to be making some sort of value judgments about how 2024 went.

9:47Some of the things I mentioned didn't go well at all for the AI industry.

9:51Some of those things

9:52actually caused huge problems in society.

9:54And it's obviously going to spawn legislation and it's going to have us discuss the concept around data privacy and security all over again in the industry.

10:02But I think this particular discussion then forces you to have to have a perspective.

10:06And this is the next most important thing when evaluating how Tableau and AI have gone in 2024.

10:12And that is, what is your perspective?

10:14What guidelines do you have for yourself when using AI?

10:17Let me share mine and maybe that will help you construct your own set of guidelines before we actually get stuck into what Salesforce's guidelines are.

10:24Like any perspective, my view on AI is largely personal.

10:28It's based on my experience of using AI over the last now year and a half

10:31And I have to say AI has been a transformational capability in my working life but also in my personal life.

10:38I've used it to do things that I just didn't have sometimes the scale or reach to be able to do.

10:43And that doesn't mean that I was using it to automate my work or using it to do things for me.

10:48A lot of the time I was actually using it as an assistive tool to sort of aid my brain in thinking.

10:53It was very much

10:54a bicycle for the mind.

10:55And that really frames my perspective because I I view AI as a tool.

10:59I view it as something that I should use to empower me and I view it as something

11:04that I should choose to wield at the right time.

11:07And so I have a bunch of principle, a bunch of guidelines that help me decide when and how to do that.

11:11And so I want to share that with you because I think it's important to have this

11:15When you're building a tool that millions of people are going to build and we'll shortly look at Salesforce's view, I want to show you mine and this will help frame the critique I'll have for Tableau AI a little later.

11:24Now these are quite simple, so I'll go through them quickly.

11:26The first point I want to touch on is purpose.

11:28I genuinely think that all AI tools need to have a genuine purpose in our lives

11:33And by purpose, I'm really talking about AI offering something more than what can already be done by an individual.

11:40A good example is writing, where you will see people using AI to write articles on the internet.

11:44And actually a lot of what we're consuming, especially on LinkedIn and on websites, I think is being generated by AI.

11:50And you can just tell there's there's some sort of

11:52standard conclusion that these things come to.

11:54There's there's a certain sort of metronomic tempo that people have suddenly started to hit with content.

11:59And you can tell that's being powered by AI.

12:01And what that does is it reduces the connection that we have with each other.

12:05And in that instance, I think the purpose of AI should be to stay away.

12:08And it should be to help us

12:10think better, think more clearly, evaluate concepts that we haven't thought of.

12:14And that's very much sort of a key thing for me.

12:16Using AI to empower what I do and the purpose should be to enhance and go beyond what I was previously capable of

12:23while still putting me in the driving seat.

12:25And that's sort of a a very hard thing to describe, but you'll see how that plays out when we start to look at some of the features inside of Tableau.

12:32Now the next one is ethics.

12:34Ethics is pretty straightforward to understand.

12:36We're talking about transparency, we're talking about privacy, we're talking about a whole range of concepts that really talk about how we view individuals on this planet if I sort of make it a larger plan.

12:46concept now every society every culture will have a different set of ethics and your business will have a different set of ethical boundaries that it will operate under compared to another business

12:56But the point here is is that when you look at the companies you're using and you're looking at the tools you're using, I think it's important to evaluate their ethics because AI is one of those tools that has the ability to

13:06um take a set of ethics and imprint it on your work.

13:09So if I go ahead and take a piece of text and I tell OpenAI or Chat GPT to write it better

13:15When it writes that context, the ethics that it uses to sort of comprise that text, to build that text up, are going to be based on the model and the way that it's built.

13:24And the ethics that goes into that.

13:26are going to touch on things like copyright.

13:28You might have a view and opinion about how work is valued in society.

13:31That is an ethical concern.

13:33You might have a view about how certain communities are represented and how they're represented specifically

13:38in literature that is again another ethical thing to be understood and so when we talk about ethics it's not easy for me to say here are a bunch of guidelines to follow

13:47What I really mean here is how do these AI tools support and enhance your own set of ethics or do they go against them?

13:55Do they use data that's maybe not theirs?

13:57Do they scrape information that they maybe should have paid for?

14:00And do they maybe put certain communities under harm?

14:03That's a really important and hard subject to touch on.

14:06I won't go into it in this video

14:07But I think it's an important one and you know what I mean.

14:09Now the last one is a is a very general one.

14:11I'm gonna just call it good karma.

14:13And this is a hard one to describe.

14:15I'm just looking at my notes here because I'm trying to I'm trying to find a better way to frame it

14:19When I talk about good karma, I think of tools sometimes as citizens in in our world.

14:24So a good example.

14:25When we drive our car day to day,

14:27I think when people buy a car, they don't just think about its ability to go from A to Z.

14:31They also think about things like fuel economy.

14:34They also think about um potentially its impact on the environment because that is a very topical thing.

14:38And so

14:39When car companies uh promote their cars, they'll talk about um those specific attributes.

14:44Because when we buy a product, what we are also buying is a perspective in some cases.

14:50Sometimes products imbue those principles.

14:52So Apple will be

14:53You know, a good example here, they have a range of products that are completely carbon neutral because they have a view, they have a a perspective on karma in the world.

15:01So their products

15:02uh are designed to try and you know alleve a positive impact on the planet.

15:07Now whether you buy that or not that's a separate discussion.

15:10But that's sort of what I'm going for here

15:12Do the companies building these models and do these models in general leave a sense of good karma in the communities that they operate in?

15:19That's a hard one to talk about, but I think it's a perspective that I can share with you as we start to look at the rest of the product.

15:25Okay

15:26Now, I think we've talked about enough about my perspective.

15:28Let's talk about Salesforce's perspective because they have some guidelines that they use to develop products.

15:34And we're going to start talking a lot more about Tableau.

15:36Okay, we're here.

15:37We're talking about AI and Salesforce and by extension Tableau.

15:41But I think it's important to just say one more thing.

15:44We often talk about Tableau like it's both a company and a product.

15:48And I think this has been a mistake in the community for a long, long time.

15:51And we're now in 2025.

15:53Tableau used to be both a product and a company.

15:56It's no longer that.

15:58It's now Salesforce the company and Tableau the product.

16:00And the reason that's important is because when we're sort of comprising our critique, I think we now have to channel it towards Salesforce.

16:08And in order to do that, we have to understand the context in which Salesforce operates in.

16:12And so what I'm going to do here is kind of split those two things apart.

16:16I'm going to look at Salesforce, their company, the way they look at AI, and then see how that translates into the Tableau ecosystem after that.

16:24So in this first part, let's look at how Salesforce has been talking about AI for the last year and a half

16:29some of the principles they have and then we can evaluate of how well those have translated to Tableau and the impact they've had on the product.

16:36AI and the Salesforce ecosystem I think is a real struggle to understand because

16:40of the way that Salesforce markets itself.

16:42I actually think a lot of the detail is there and it's very easy to find, but it's very easy to also get confused because of some of the way the products are named and some of the way the platform is presented to you.

16:52use it.

16:53So let me try and sort of synthesize this in the most basic way possible and just give you the things you need to know.

16:59The best place to start is with the Einstein One platform.

17:02Now, the way this works is that the Einstein One platform is basically the platform upon which all of Salesforce operates under.

17:09And this is a fairly new thing.

17:11Um it was clearly a move to sort of get them ready for AI, hence the term Einstein.

17:16You'll come back to that a lot in a minute.

17:18But essentially think of this platform as the foundation, the

17:22uh uh the ground upon which all the different services that sales force have operate under and I will have something up on screen you'll see that Slack

17:29Quip, Tableau, Heroku, and then all of the CRM apps operate on.

17:35And then based off that you have these sort of um two wings, two flanks as it were.

17:39You've got the data cloud where your data resides, then you've got Einstein on the right hand side.

17:44And in between those you have uh bridges, let's call them.

17:47So on the left we have real-time data, this concept of being able to access your data in situ

17:52And then on the right, you have something called the Einstein Trust Layer, a kind of bridge between your CRM apps and Einstein.

17:58And so the best way to think of this is to think of this as the following.

18:02Einstein 1 is the platform upon which everything operates under.

18:06Einstein is actually the brand that is used for all AI capabilities.

18:11The Einstein trust layer is what allows your data to live in that world, essentially creating some sort of separation between

18:18your data that should stay private and personal, and all these AI models that are being run by large operators like OpenAI.

18:26Essentially Einstein the Einstein Translate is the bridge upon which that works

18:30If we actually switch over to the screen here, there's actually a really good fact sheet that they have that kind of goes into this and kind of digs into a lot more detail.

18:38And I'm just sort of scrolling uh through it now.

18:40You'll be seeing it on screen

18:41And as we start to sort of go down the feature list, you'll see that Einstein is essentially tagged onto pretty much everything.

18:48And this should show you that anything with Einstein is essentially an AI capability

18:53But the other thing to note here is that actually in 2023 um AI had this uh I think transformational change which which meant that it basically became the headline name, the headline title for all the disciplines of

19:07previously sort of sat um in a much broader set of context.

19:11So for example, features that were previously termed

19:14uh machine learning and now called AI features.

19:18Predictive analytics, some of those uh capabilities are now starting to be termed under the AI banner.

19:23And so the other thing to remember is that

19:25AI will also incorporate a lot of features and capabilities that were previously there before, but came before this AI hype train.

19:32And so AI has become the big banner.

19:35um and it's both a discipline within that field as well as the banner title if that makes sense.

19:40So that's sort of my way of explaining this.

19:42The w the one thing to note here on this page though is that um there's no mention of Tableau and this is just two clicks

19:49Off the home page, if you search Salesforce AI, you go to a page about Einstein, you go click on the fact sheet and there is no mention of Tableau and we're in 2025.

19:58I think that's a little bit awkward.

19:59There's a still a mention of CRM analytics, obviously the existing analytic capability inside of the Salesforce ecosystem.

20:06But I think it's just strange that there is no mention of Tableau in this fact sheet whatsoever.

20:10Maybe they've forgotten to update it.

20:12But I think details like this really start to matter when we're talking about Salesforce.

20:17AI principle carrying through to Tableau.

20:19How can that even work if it's not represented here at the high level?

20:23Okay

20:24The next thing to understand are the guiding principles that Salesforce uses to build products because they actually have these and I actually think these are really, really good.

20:32They're really hard to find though, so what I want to do is sort of surface them to you and show you what they are

20:38So Salesforce have this concept called Guidelines for Responsible Development.

20:42There's a blog post which you'll see on screen right now.

20:44And in that blog post back from February 2023.

20:47They outline these sort of core principles and I'll put them up on screen.

20:47.

20:48.

20:51Essentially, they talk about accuracy, safety, honesty, empowerment, and sustainability.

20:56In a funny way, these align very much with some of my own personal perspectives that I thought about earlier.

21:01It's obviously termed slightly differently and different things go into the different headers, but it's very clear that they've put a lot of thought into it.

21:06I'm not going to read them out now.

21:07I'll I'll leave a link to these in the description so you can go check them out.

21:11But what I found interesting is if you dig into these, there's actually a lot more behind it.

21:15There's a lot more thought that's gone in there.

21:16And there's another sort of set of uh information that you can find here when you search trusted AI in the Salesforce ecosystem

21:22And it talks about their commitment, those same values, and how they work around this concept of trust, which I think makes a ton of sense.

21:30And then there's also a ton of transparency about the way their models are built and some of the ways their models are trained

21:36The issues and challenges.

21:37If you go into some of these model cards, there's some really detailed information about how their models could do better, how they could be improved

21:44Tied into research that's being done in the community about those specific things.

21:48So if you really care about sort of this foundation and understanding in terms of ethics, but also understanding how you make sure that your customers

21:57aren't abusing these uh principles in order to do things in their companies that could potentially harm people.

22:03They have they have something called the AI accessible use policy.

22:07And in that policy, it actually restricts customers from using their services from misusing AI, which I think is actually really, really good.

22:14Again, it's hard to find, but it's there and it shows there's actually a lot of thought has gone into it.

22:19And so this makes me feel sort of fuzzy and warm inside because on one hand I actually think, yeah, you know, I had a bunch of views and perspectives and uh and and thoughts about how AI should be used and it turns out that at least Salesforce

22:31does have this thought, it has it documented, it has it easy to share, not easy to find.

22:36I think they could do a lot of better.

22:38But Salesforce has that information.

22:40Now the challenge starts to come when you start to think well how does this translate to Tableau?

22:44And how do the features specifically start to resonate with some of the other principles we talked about earlier in the video?

22:51It's now time to talk about Tableau and AI.

22:54Let's get into it.

22:55So when it comes to Tableau and AI, it's an interesting story because actually quite a lot of features have existed for some time before AI was a bit of a hype train.

23:03And so if we think about explained data in our stata, these would be features that today come under the AI banner, but have actually existed before the hype.

23:11And so you can kind of see these as part of that journey that Tableau had already started and already been on.

23:16They use capabilities such as machine learning and statistical methods in order to help analysts, but fundamentally they fall under this larger banner.

23:23And then came features like Tableau Pulse.

23:26Now Tableau Pulse is sort of interesting because it's

23:28largely grounded in statistical methods in order to give you the insight that it generates, but it uses generative AI to summarize those insights.

23:36So

23:37The actual AI bit here isn't the analysis, it's the summarization, the ability to send you that email and give you a hint of what's going on across the landscape of metrics that you follow.

23:47So Tableau Pulse was really the first ground up capability that was built from scratch in this new area.

23:53And then following that we had the Tableau Agent.

23:56Now the Tableau Agent is interesting because

23:58It's a feature, but it's spread across multiple products.

24:01So you'll have it inside of Tableau Prep Web Authoring Experience, and you'll also have it in the Tableau Authoring Experience again in the web.

24:08So these features only run in the web.

24:10And I actually did a survey of users and I asked everyone on LinkedIn, hey, look, when you build something, um, whether it's in Tableau Prep or in in Tableau in general.

24:19Do you start in the desktop or do you start in the web?

24:21And actually the responses came back predominantly for desktop.

24:24And this sort of starts to point at some of the challenges I see.

24:28A lot of this new world is designed for the web first.

24:31And I think that's largely because of a challenge around legacy in the product you see

24:35AI is a fairly new thing and in order to incorporate it, you'll see that Tableau Pulse was a net new build.

24:41You can you you even see that when you go to the Tableau Pulse experience.

24:45It has a very different vibe.

24:46It definitely is a sort of new look tableau compared to the traditional legacy tableau that we've seen.

24:51And that was actually one of the first features to involve AI.

24:54We've then had Tableau Agent, which took a lot longer to implement into the platform.

24:59But you see they've only done it in the web.

25:00And so this is the biggest hint.

25:02All of the AI capabilities have only really come through in the web interface.

25:06Now

25:07This is both a good thing because it shows that Tableau is able to push these features through much, much faster in the web authoring experience compared to the sort of legacy desktop setup.

25:15But it also does start to sort of pose some questions, which I'll come back to later on about things that I think are missing in Tableau's journey involving AI.

25:24But anyway, let's stick, let's stick to the sort of existing points that we're on here.

25:27Now the other thing about the Tableau Agent is that it has different capabilities in different products.

25:33So in Tableau Prep and in Tableau Web Altering, you can get it to help you with calculations.

25:37And this is great.

25:38because it actually helps people who are trying to write calculations or know what they need to write but just want to save a bit of time and you can go in and use it.

25:45In the web altering experience, especially for authoring

25:48You also get this other ability to allow you to build charts.

25:51You can ask it to do some analysis for you and it will go ahead and use the interface and build it for you.

25:55This is a great way to learn, there's a great way to understand it.

25:58But again, it's only available through web authoring.

26:01And so

26:01What we're starting to have is a a body of features that are starting to sort of pop up in disparate places.

26:07But the big challenge is you can tell

26:09is that there's no one cohesive experience.

26:12You know, Tableau Agent is very much limited to the side pane.

26:15It's sort of very much off to the side.

26:17And what is evident from all of this is that actually the biggest challenge Tableau have

26:21is getting AI into what feels like a modern interface and the current sort of legacy setup that we have is no longer suitable for this.

26:30Now

26:31There is another argument you could make here.

26:34Earlier on when I talked about some of the trends in 2024, it was very obvious that one of the big transformational sort of steps in 2024 was the ability to run models locally.

26:45This idea that you don't need the cloud to run the model, you can actually get these models scaled down to fit on your laptop or even be able to run them on the edge.

26:53And so what is interesting is that while Tableau is pushing forward with this sort of cloud-first mentality

26:58There is an emerging move in the industry to do the opposite.

27:01And so we haven't heard much of this narrative.

27:04And actually what's interesting is that we've not even seen Tableau deliver or Salesforce deliver.

27:09on their promises to allow us to use different models inside of the AI product.

27:14Way back when when we first saw uh this sort of

27:17first instantiation of the Tableau agent in the cell source ecosystem, there was talk about uh allowing us to use different models to do different things.

27:25And in today's world that would just be like using Chat GPT for one task, using Claude for a separate task

27:30And then using Google Gem Knife for something else.

27:32Essentially being able to choose your foundational model inside of the Tableau ecosystem for various reasons and then being able to build on those to perform various tasks.

27:41Now that that hasn't transpired

27:43It's also not very clear what version of ChatGPT that Tableau are using behind the scenes because as I understand it, everything is currently working with OpenAI.

27:53But

27:54Uh just from my day-to-day use, it's very, very obvious it's not the latest and greatest version of OpenAI that I have on my laptop and that you can use on the web service.

28:03It feels like an older version.

28:05So the the transparency there is sort of lacking.

28:08And then the ability to sort of have those options to switch out and the modality as it were between models.

28:14is actually something that the industry decided in 2024 was a good thing to do.

28:18A lot of tools allow you to switch between different models right in the same interface in a very simple way.

28:24We've not seen any of this inside of the Tableau ecosystem.

28:26And then on top of that

28:28If you're on Tableau server, you're not getting any of this AI benefit.

28:32And so the big challenge here is while the industry is talking about having edge AI and having m models run locally

28:38I don't understand why there couldn't be a discussion about being able to run models locally on Tableau Server.

28:45Yes, you might have to tell Tableau Server Admins to really beef up their servers.

28:49But in a world where I can run Lama 38, I think it's 48B 3.

28:542 on my laptop, I don't see why a company couldn't decide to say, well, actually this technology is useful.

28:59We want to run it locally on our Tableau server to do the kind of summarizations that are happening here on this platform.

29:05Tell us the specs of the server we need

29:07and we'll make sure we set up our servers for that.

29:10Some companies are doing that.

29:11A lot more companies are, you know, going back off the cloud onto on-premise infrastructure.

29:15And I think it's a trend that we'll carry on through.

29:18And we're not seeing any of that discussion at all inside of the Tableau because

29:21system.

29:21So that to me is also red herring given that it was such a big trend in 2024.

29:25So everything I've touched on so far is architectural.

29:28It's it's all about the high level design of the entire Tableau ecosystem.

29:32And it's actually quite a poignant time to talk about that because as you know, Tableau have been marketing the next generation of their platform.

29:39Now one of the things that drives me nuts is that the marketing talks like this product has already been launched

29:44If I go onto the Tableau feed, I'll see a post pretty much once or twice a week showing us uh uh this sort of a new thing that's coming.

29:51And the problem is we started talking about this new thing that was coming.

29:54back in April and we're now in January of 2025 and we don't have access to it and so um it's very hard to talk about this because no one really knows but

30:03What must be clear here is that as this new vision for Tableau comes about, it's clearly going to be the platform upon which AI is a really prominent capability.

30:14Um, we're seeing capabilities inside of what I would now call the legacy products today um coming through as they were.

30:20I would also argue that actually

30:22unbeknownst to I think the community in general, I think the number of features and the pace of features coming through is much higher.

30:29I'll put some slides up to show you.

30:31some of the features that are coming through from the community and just the sheer number of features over time that are coming through in the most recent releases.

30:37And I think that's an important thing to call out.

30:39But those features are not necessarily in the places that necessarily the whole community wants

30:45And so as we go forward into this new tableau with agent force or tableau einzeln, whatever they decide to call it, as we go into this new world

30:52I think that platform is likely to be the place where AI is a prominent capability, which will then, I think, make some of these older

31:01platforms these older setups of Tableau a little bit sort of luckluster especially if people are wanting that AI capability so that's an important thing to talk about from a high level.

31:10Now if I go back to my general principles

31:13that I had.

31:13I think it was purpose.

31:14Um looking at my notes, empowerment, ethics, and good karma.

31:17We look at purpose.

31:19I think the tableau agent is one of those things that I really really struggle to understand

31:23You see, on one hand, I actually really like the term copilot, and this is not not necessarily something that I'm attached to, but I think it was a meaningful description of what was happening inside of the product experience

31:34you're writing or you're building something and you want some support, there's a copilot that allows you to go and get that task done.

31:41Now when we talk about agents in the Tableau ecosystem, especially when we talk about agentic AI

31:46and the meaning of a gentic AI, which is very much an autonomous system doing things on your behalf, I struggled to translate that to the Tableau ecosystem.

31:54This is one of those places where I wish um Tableau as a team or Tableau as a product team inside of Salesforce

32:00Could have just stood up a little bit to the Salesforce messaging and said, actually, we haven't fully delivered on the basic AI experience just yet.

32:08We are a way off from agentic AI.

32:11Let's just stick with copilot for a little bit longer.

32:14Let's actually deliver on some of that co-pilot excellence.

32:16If you use Tableau Agent today, it was previously called Einstein Copilot.

32:20I don't think that vision has been fully delivered.

32:22And so I think

32:23It was a necessary thing to finish that vision first before then trying to switch up the game and then focus on the gentic AI because the thing is.

32:31.

32:32The marketing about agentic AI talks about autonomous systems going off and doing things behind the scenes.

32:38Now that makes sense in the Tableau Agent Force, Tableau Einstein world, but in the

32:42In in in today's world, every single day as we're working in there and that side panel, when I go in there, that to me is not an agent.

32:48That's still very much a copilot.

32:50And it makes that experience a little bit painful because it it it makes it sort of much further from the reality that we're heading to, if that makes sense.

32:59And I'm sure there's lots of work going into making that feature much, much better in 2025.

33:03But as of where we left it in 2024, in the term Tableau Agent with a GenTic AI, I think that's really, really far from the vision that's being delivered.

33:13Now, the other thing is that when you look at that agent, in some cases it does really well.

33:18And this is the thing I've been struggling to sort of battle with in my mind.

33:21When we look at Tableau Prep, I think it does a great job because

33:24You're in there writing a calculation, you need some help, boom, it can go off and help you think about that calculation.

33:29And even if it doesn't get it right, it's okay because the context is so small.

33:32You're just writing a calculation and you can see the output, you can see what's going on, you can see the way that it's all done.

33:39However, when you're inside of web authoring and you start to ask it for help,

33:43I just get it misfiring a lot more on even basic things.

33:46Even basic things like understanding the capability that's available in the product

33:51Even using the pre-canned responses they try and get you to use as part of a tutorial, it's just not firing on all cylinders.

33:58And so there is something sort of disparate about this too.

34:01And I think it really boils down to this point around empowerment

34:04I think AI when you use it, when you get it to do something on your behalf and you arrive at the answer without seeing all the steps that took you there, it it kind of uh has this sort of visceral disconnect with the outcome.

34:16It's a bit like IKEA.

34:17When you buy IKEA furniture,

34:18Part of the value in IKEA furniture isn't actually in the price, isn't actually in the quality.

34:23It's in the very fact that you build the furniture yourself.

34:26So once you've built the furniture, the sense of pride, a sense of accomplishment.

34:29And it's because you were part of that.

34:31Now, IKEA could very easily sell you complete furniture and just ship it to your address.

34:36That would be easier than owning a store.

34:38It'd be a lot cheaper, and they could just ship it directly from where they get to.

34:41But here's the thing

34:42That's not part of the experience.

34:43That's not the point.

34:44The point is the very connection you have with making the product.

34:48That is the whole IKEA brand

34:50And so it it's sort of an interesting thing when we start to talk about AI here where we have this agentic um AI, which I think is going to start to um

34:59fragment the connection that people have with data, which is sort of counter to the point.

35:04If you look at Tableau, Tableau is very much an ecosystem.

35:07that it prides itself on people crafting beautiful stories.

35:11If you look at the Ironviz, which has just had uh some feeder announcements recently, if we look at Tableau Public, those are all big marketing pushes to really tell the story that look

35:22tablea is about crafting stories around data.

35:25That is very much sort of the underlying principle behind those platforms.

35:28And then here with agentic AR, we're kind of doing the opposite.

35:31We're kind of

35:32creating a disconnect between uh that narrative and how you get there.

35:36And if you get there without understanding how anything is done, how can you tell the story in in a good way?

35:41And so that's where I really struggle.

35:43In Tableau Prep I love it

35:44In web ordering, I don't because I think it creates this disconnect.

35:48Whereas in Tableau Prep, because it's only doing sort of small chunks of work, I think it's a little bit better.

35:52So that's sort of

35:53Um a small point about purpose there.

35:55And the other thing is, you know, if you're using these agents, I think they can create a new problem.

35:59And this problem doesn't exactly help the situation, which is

36:02The need to check and verify what's being done.

36:05Especially with around new analysts, if they're using this capability and you know they're using it, I think it does increase the burden on me as someone who's trying to educate them to have to check more of their work.

36:15Whereas I think if I

36:17Set off analyst and say, look, let's give you this path that you can follow.

36:21It doesn't involve AI, but I know that as you go through those steps, you're gonna learn a couple of core principles and I'll see that play out in your work.

36:27So you won't be trying things that are advanced

36:30well before you're ready to do them.

36:32The problem with AI is that you can go in there and on day one you can ask a question that you don't know is an advanced concept, but it will do the question for you and build it and give you an answer.

36:41But you don't have the skills to really validate and check that that answer is correct.

36:46And so that's that's sort of a new problem we have.

36:48I don't think there's a perfect answer to this.

36:49I'm not saying we should avoid using AI for this specific reason

36:53But I think it's a new problem and that new problem doesn't necessarily help the already bur big burden we have on being able to validate and check data sources that have taken time to build lots and lots of trust with end users.

37:06It takes one wrong answer for someone to lose trust in a data set.

37:10It takes years of work as a data team to build that trust with your consumers.

37:14And my worries with AI

37:16There's a good chance that sometimes that will break.

37:18So a big thing I'd love to see is a bigger focus on empowerment.

37:22Help me get to the answer faster.

37:24Help me build those advanced solutions faster.

37:26But when I ask a question, instead of just doing it for me, maybe teach me how to get to the answer.

37:32Step me through a bit like you would in Chat GPT.

37:35Step me through the steps so that I understand all the key concepts.

37:38That's why I think it's really interesting.

37:39Andy did a video about how uh Google LLM

37:43uh is able to guide you through how to do something in a video.

37:47I'll also I'll also made a video about this later next week but

37:50Nonetheless, I think this is really powerful because it's it's actually the learning element that I think is really really valuable with AI rather than the ability for it to be autonomous and do things behind the scenes.

38:00Anyway

38:00I think I've murdered this point.

38:02The next point is around ethics.

38:04Now I think the ethics are good here.

38:05Salesforce's ethics around how they use AI are pretty good.

38:13There's this issue around copyright and the approach that these large models have made to go and get data that I think

38:22is worth um worth exploring.

38:23Now Salesforce is kind of immune to this a little bit because um they didn't go and do this and build their own model.

38:29Maybe they're doing that behind the scenes.

38:31Um but the large companies, all of them that have built models, even Apple and a a company that I hold to very, very high standards, have done this.

38:37They've essentially gone and scraped a ton of information from the internet.

38:40before then asking for permission to actually use that data.

38:42And so when we have that kind of capability in our product, I think there is a an ethics question around sales as a company.

38:49making sure that it supports the sources uh that have created some of this original work through its other initiatives.

38:55And I think that's something that's really hard to sort of pin down.

38:57Like where's the right place to talk about that?

38:59It's probably in the largest Salesforce organization

39:01But I think it's still important because we're out here using this AI inside of our products.

39:05But nonetheless, I think it's important to value how that information and how that AI um model was created in the first place.

39:13Okay, so it's a couple of days later.

39:15Yes, I've put the same shirt back on so I don't disrupt the flow of the video.

39:19Um, but I realize I didn't talk about pricing, and pricing is actually really, really important

39:24The reason it's important is because I think there is a point around marketing that comes back to pricing, that comes back to the way we use the product.

39:32So bear with me as I try and sort of um tell the story, but I think I think it will make sense by the time I get to the end of it.

39:39So just give give me a couple minutes here.

39:41So

39:41The first thing to realize is I mentioned a survey I did on LinkedIn and what is pretty clear and I think is is still fairly true is that most users are still bound

39:51to the desktop product.

39:52Most users who are creating assets are still bound to desktop products.

39:56So when we talk about AI specifically and we talk about the features that Tableau have brought, those are all based in WebAuthring.

40:03Now

40:03That's not to say that they're not valuable, but when you put that capability into web authoring, what you have to ask is, well, what are people doing in web authoring today?

40:13that makes sense.

40:13And if you look at that survey, I'll post a link to it again, you'll see that most people are going there to sort of complete work and maybe make a couple of changes.

40:21But it's very rare that people are genuinely starting new experiences in that part of the product.

40:26And so if I just bring up the Tableau pricing page and I look at this pricing schedule, I'll do a separate video on this because this changed last year and I kind of missed the plot on this, but

40:35There's essentially three prices.

40:36There's the Tableau, there's the Enterprise and Tableau Plus.

40:38Now, Tableau is probably for no one.

40:41Like Tableau is probably just there as an entry product.

40:44But enterprises really what they want you to have.

40:47It's that sort of marketing trick where you have three options and really the one in the middle is actually the one.

40:51um that you shouldn't go for and the one at the top is the one they want you to get but nevertheless um ai is only available in the tableau plus option now when you click on the tableau plus option

41:01It sends you to another page where you have to send in some detail.

41:04So if I go back to my values that I mentioned earlier, in terms of transparency, it's actually not clear what price you pay to have the AI capability.

41:12It's interesting that all the other functionalities very

41:15you know, clear up front, you can play around with these two options and see a clear difference in price.

41:19What that suggests is that the price for AI is is probably something more fluid.

41:25and is maybe on a customer to customer basis predicated on things like your size and your clients, but there is no sort of standard set price.

41:33um that's available there.

41:35Another thing I always find interesting, look, there's a big fat 20% discount which kind of tells you where their margins sit in terms of um where they're pricing the product.

41:42But nevertheless, I'm not a pricing expert, I'm not a pricing analyst

41:45So I'm not going to sit here and say whether this is a good price or not.

41:49But this is the most important thing.

41:52When you then go back to the AI capabilities and go back to the earlier point I made around curation.

41:58It's very clear that Tableau is a marketing function.

42:01The marketing functions in Salesforce, when they market Tableau, they market what I would like to think of someone called an artisan.

42:08Now I'm actually borrowing phrases from Maltrix.

42:10This is this is a tool that's

42:11Not doing too well at the moment in my honest opinion, but nonetheless an artisan is uh someone who is essentially

42:19um you know someone who is crafting something beautiful making something that lots of people are going to to use and enjoy but when you actually look at the

42:30capability that they sell.

42:32If you look at who they talk to, if you look at the audiences that they are selling to, if you look at the Gartner Magic Quadrant, if you look at all of that.

42:39The conversation there is actually about the curator.

42:42The curator, the person who's building these governed data sets, who's managing the security, who's

42:47uh you know uh enabling other people to uh do things.

42:51So when when Tableau is selling the product they talk about curators and when they are

42:55uh marketing the product they talk about artisans right so this is sort of the the the the two uh visions that you t you t you tend to sort of see.

43:04And where this where this sort of causes an issue is there's a sort of an unspoken role for your creator in that discussion.

43:10Now the curator is something that hasn't really

43:12existed much in the Tableau ecosystem.

43:14I wouldn't say creator roles are curators.

43:17A curator is very different to a creator.

43:20A creator is someone who builds something, deploys them, and gets them out.

43:22A curator is someone who then

43:25views all of that content from a perspective of understanding how it's made and decides what is good, what goes where.

43:32And in the Tableau ecosystem, there's no role that recognizes that.

43:35It's not the server admin.

43:36That's

43:37That's a role that's it's purely designed to sort of get the structure and get it all working.

43:41Essentially think of a server admin as a mechanic on the platform, making sure it's working, it's working well, security's done well.

43:47A curator is something very specific.

43:50And actually

43:51In the new vision for Tableau, the curator is actually one of the most important roles.

43:56That's going to be the person who, you know, puts together semantic models.

44:00that puts together workspaces that you know curates and guides people and makes sure that the things they need in their day-to-day work is actually available where where it should be and they'll be probably delegating work to creators

44:12in this sort of new vision that we're talking about.

44:14And so it makes me ask this question, and I'm getting to my point, just please bear with me.

44:18I'm getting to my point.

44:19So it makes me ask this question, who's the customer and who's the user?

44:24Because when you think about the pricing, I think the pricing talks to the customer.

44:29The customer being the executives, the customer being um the people who actually pay for Tableau, the people who are at the end of the invoices there, who are looking at the purchase orders for Tableau.

44:39But the user is not the same person.

44:42And so a lot of the AI in my opinion, I think, targets the customer, the

44:47um the person who's actually paying the invoices and is maybe pitched towards a world where the curator is a more prominent role.

44:54But in today's world, in the products that we use, there is no such thing.

44:58It doesn't yet exist.

44:59There's no capability that's designed around it.

45:01It's very much something that's coming in the future.

45:03And so we end up in this, I think, difficult place where I think the pricing reflects the future of where Tableau's heading.

45:11Hence you've got these sort of demarcated Tableau Enterprise and Tableau Plus

45:15It very much tees up a role for curators in the future, but we're not quite there yet.

45:21And so the pricing

45:23No, it it it it makes sense.

45:24I again I'm not a pricing analyst, so I can't tell you whether this is good or not.

45:27I've never used other tools and seen the invoices, so I can't tell you whether these are cheap or good or whatever.

45:33I know that's the obvious thing to sort of

45:35decide and you can you can maybe post in the comments if you've paid for one of these and you think they're great value.

45:39The people I've spoken to who have access to Tableau Plus do think it's great value for lots of different reasons, okay?

45:45So I'm not here to say that.

45:47What I am here to say though is that the pricing is an important part of this narrative.

45:52Um just yesterday, a couple of the large AI companies, Microsoft, Google, um, they've all done something

45:59fairly bold which is they've said to all the existing customers look I know you didn't ask for AI but we're changing our pricing model we're gonna add a few more dollars onto your price for you because we think AI is now part of everyone's work

46:13even though you didn't ask for it.

46:15And um the price is going up for you.

46:17And then they've gotten rid of their top tier, their AI tier

46:21I'm wondering if this pricing setup on Tableau website is actually the middle step towards that.

46:26So when Tableau Einstein comes along

46:29you'll have uh no need for the tableau base model here and essentially you'll just have the enterprise and the tableau plus and then tableau plus might have levels

46:38that have AI all the way through this.

46:40So enterprise will actually get the Tableau Plus capabilities and then Tableau Plus will maybe be something more advanced.

46:45Or maybe it will just disappear.

46:47But it definitely feels like that's where a lot of companies are going to head in the future.

46:50Microsoft have just done it, Google have just done it.

46:51I'll try and put links to that in the post.

46:53So it's a it's a it's an evolving topic.

46:57I can't sit here and give you an opinion whether the pricing is good or not.

47:00Again, I've told you why, but it's definitely strange because I think the pricing speaks to one type of person in the ecosystem.

47:07But the marketing, the marketing of the product speaks to a completely different person, and that person isn't reflected here in the pricing.

47:13Anyway, those are my thoughts back to the video

47:15The very last point is about good karma.

47:17Now I think the good karma point is actually quite easy to cover here inside the tablay because I genuinely think that again Salesforce and the product managers are putting a ton of great thought

47:27into the product to make sure that it really delivers on all the promises that everyone has.

47:32Now this is going to be something that will evolve over time.

47:35So it's not easy to just say look here's where we are in 2024.

47:38I think it was good or bad

47:39I think it's easy to actually find an issue maybe next week that makes us reflect on 2024 and think that actually 2024 wasn't good.

47:47So this is a moving target.

47:49But I think based on what I showed you earlier on from Salesforces Ethics

47:52and how they're going about principles of developing these features.

47:56And then what we're seeing about the future of Tableau, there's a lot of transparency here.

48:00There's a lot of detail.

48:01And I think the best way to really frame this discussion is actually to become part of the community.

48:07And so this leads me to my very final point for this video.

48:10Come and get involved in a community panel that's happening in a couple of weeks.

48:13I'll be moderating along with a bunch of other hosts.

48:16I'll put them up on screen here.

48:17I can't go through all of them right now

48:19But there's going to be a great discussion about this topic.

48:22And I think we'll cover some of the topics that we've discussed in this video, but we'll also cover some of the new concepts that I think everyone in the community is thinking about.

48:30So go ahead, get involved, and I'll catch you in the next one.